Nomenclature: Flexible Name Resolution on a Flat Object Namespace
With the introduction of storage class memory, we have the opportunity to change how data is represented. Currently, there is a clear separation between volatile and non-volatile storage, largely represented by the file system and block device drivers. This separation will cease to exist when SCM enters the memory hierarchy. We expect data to be stored in transient objects, making data storage more akin to a flat object store than a hierarchical file system. For human convenience and backwards compatibility we will want to name and classify these objects. We will do this through flexible name resolution using programmatic resolution of objects. Names can range from complex search queries to a simple one-to-one mappings from a string to a GUID. Due to the simplicity of the Twizzler object store, the name resolution methods of Nomenclature let the same set of objects can be found in multiple ways in less steps than conventional indexing systems. Nomenclature can take the form of a hierarchical file system lookalike, or a collection database, where objects are resolved by associated metadata or content rather than a simple name and is fully extendable by additional resolver modules. Nomenclature gives flexibility over data access and organization, and will enable transient objects to be effectively assessed on future systems.
Monday, January 22, 2018 at 12:00 PM
Last modified 19 Jan 2018